1 Introduction
2 Hamiltonian systems
-
Energy is preserved: \(H({{\varvec{x}}}(t))\) is constant in t.
-
For every t the mapping \({\varvec{\varphi }}^t\;:\;{\mathbb {R}}^{2n}\rightarrow {\mathbb {R}}^{2n}\) is symplectic (or canonical), that is, its derivative is a symplectic matrix at every point.
-
The mapping \({\varvec{\varphi }}^t\) is time symmetric, i.e. \({\varvec{\varphi }}^{-t}({{\varvec{x}}}(t))={{\varvec{x}}}_0\) for every t.
3 Symplectic subspaces and low dimensional approximations
4 Exponential integrators
4.1 Exponential Euler method
4.2 Explicit exponential midpoint rule
4.3 Implicit exponential midpoint rule
5 Forming the local basis using Krylov subspace methods
5.1 Equivalence of the Krylov and the local system approximations
5.2 Symplectic Krylov processes
-
Now \({{\varvec{H}}}\,W_k({{\varvec{H}}},{{\varvec{v}}})\not \subset W_{k+1}({{\varvec{H}}},{{\varvec{v}}})\), generally.
-
If p is a degree \(k-1\) polynomial, then \(p({{\varvec{H}}}){{\varvec{v}}}\in K_k({{\varvec{H}}},{{\varvec{v}}})\subset W_k\).
-
Arnoldi in \({\mathbb {R}}^{2n}\): the approximation is not Hamiltonian.
-
Hamiltonian Lanczos: breakdown, early loss of symplecticity.
-
Isotropic Arnoldi: does not include a Krylov subspace.
-
Symplectic Arnoldi: expensive.